CN109951724A

CN109951724A - Recommended method, main broadcaster's recommended models training method and relevant device is broadcast live

Info

Publication number: CN109951724A
Application number: CN201711386148.3A
Authority: CN
Inventors: 肖蒴
Original assignee: Alibaba Group Holding Ltd
Current assignee: Alibaba Group Holding Ltd
Priority date: 2017-12-20
Filing date: 2017-12-20
Publication date: 2019-06-28

Abstract

This application provides a kind of live streaming recommended methods, device and system, propose a kind of new live streaming way of recommendation, and the angle recommended from main broadcaster is that user selects main broadcaster to be recommended, are watched with the related live video for facilitating user to select oneself interested main broadcaster.In addition, for obtaining mathematical model using the image training comprising characteristics of human body of main broadcaster, which is used to select main broadcaster to be recommended for user present invention also provides a kind of main broadcaster's recommended models training method and device.

Description

Recommended method, main broadcaster's recommended models training method and relevant device is broadcast live

Technical field

This application involves direct seeding technique field, in particular to a kind of live streaming recommended method, a kind of live streaming recommendation apparatus, one kind Recommender system and a kind of main broadcaster's recommended models training method and device is broadcast live.

Background technique

With the development of network video technique and the more diversification of video content, there is the view of various different fields daily Frequency class program appears in the visual field of numerous online friends.The online habit of more and more online friends via traditional word read or is seen Filmlet, is transformed into that this expression of viewing live streaming is more intuitive, content is richer, the more real-time program category of interaction.

As live streaming operation platform, it is necessary to solved with more scientific method how user stop shortest time in The main broadcaster that user recommends its most interested, to meet the individual demand of user, this is a crucial function for influencing user experience Energy point, and influence the key factor that user watches frequency and duration.

Summary of the invention

In view of this, proposing the new live streaming of one kind this application provides a kind of live streaming recommended method, device and system and pushing away Mode is recommended, the angle recommended from main broadcaster is that user selects main broadcaster to be recommended, to facilitate user to select oneself interested main broadcaster Related live video watched.

Present invention also provides a kind of main broadcaster's recommended models training method and devices, for utilizing the special comprising human body of main broadcaster The image training of sign obtains mathematical model, and the mathematical model is to select main broadcaster to be recommended for user.

A kind of live streaming recommender system is provided in the application first aspect, comprising:

Server-side generates main broadcaster's recommendation list of the user, the main broadcaster for selecting main broadcaster to be recommended for user Recommendation list is used to record the relevant information of the main broadcaster to be recommended, sends main broadcaster's recommendation list to spectator client；

Spectator client, the main broadcaster's recommendation list sent for receiving the server-side, recommends according to the main broadcaster List is that user shows live streaming relevant information.

A kind of live streaming recommended method is provided in the application second aspect, this method is applied to server-side, comprising:

Main broadcaster to be recommended is selected for user；

Main broadcaster's recommendation list of the user is generated, main broadcaster's recommendation list is for recording the main broadcaster's to be recommended Relevant information；

Main broadcaster's recommendation list is sent to spectator client.

A kind of live streaming recommended method is provided in the application third aspect, this method is applied to spectator client, comprising:

Main broadcaster's recommendation list that server-side is sent is received, main broadcaster's recommendation list is used to record the phase of main broadcaster to be recommended Close information；

It is that user shows live streaming relevant information according to main broadcaster's recommendation list.

A kind of live streaming recommended method is provided in the application fourth aspect, this method is applied to main broadcaster's client, comprising:

The commodity classification recommendation list that server-side is sent is received, the commodity classification recommendation list is for recording concern main broadcaster User group end article classification；The end article classification of the user group is according to the user group to commodity Historical operation behavioural analysis obtains；

It is that main broadcaster shows commodity classification according to the commodity classification recommendation list.

A kind of live streaming recommendation apparatus is provided at the 5th aspect of the application, described device is applied to server-side, described device Include:

Selecting module, for selecting main broadcaster to be recommended for user；

Generation module, for generating main broadcaster's recommendation list of the user, main broadcaster's recommendation list is described for recording The relevant information of main broadcaster to be recommended；

Sending module, for sending main broadcaster's recommendation list to spectator client.

A kind of live streaming recommendation apparatus is provided at the 6th aspect of the application, described device is applied to spectator client, described Device includes:

Receiving module, for receive server-side transmission main broadcaster's recommendation list, main broadcaster's recommendation list for record to The relevant information of the main broadcaster of recommendation；

Display module, for being that user shows live streaming relevant information according to main broadcaster's recommendation list.

A kind of live streaming recommendation apparatus is provided at the 7th aspect of the application, described device is applied to main broadcaster's client, described Device includes:

Receiving module, for receiving the commodity classification recommendation list of server-side transmission, the commodity classification recommendation list is used In the end article classification of the user group of record concern main broadcaster；The end article classification of the user group is according to the use Family group obtains the historical operation behavioural analysis of commodity；

Display module, for being that main broadcaster shows commodity classification according to the commodity classification recommendation list.

A kind of main broadcaster's recommended models training method is provided in the application eighth aspect, this method:

Obtain the image comprising characteristics of human body of main broadcaster；

Detection described image obtains the key point of characterization organization of human body feature；

The corresponding feature vector of each main broadcaster is calculated according to the corresponding key point of each image；

Main broadcaster's recommended models are generated according to the feature vector of all main broadcasters.

A kind of main broadcaster's recommended models training device is provided at the 9th aspect of the application, which includes: acquisition module, is used In the image comprising characteristics of human body for obtaining main broadcaster；

Detection module obtains the key point of characterization organization of human body feature for detecting described image；

Computing module calculates the corresponding feature vector of each main broadcaster according to the corresponding key point of each image；

Generation module, for generating main broadcaster's recommended models according to the feature vector of all main broadcasters.

Compared with prior art, the application includes following advantages:

Main broadcaster's recommended models training method provided by the present application and device, using the image comprising characteristics of human body of main broadcaster as Training data detects the key point of the characterization organization of human body feature in the image of each main broadcaster based on these training datas, The corresponding feature vector of each main broadcaster is calculated based on these key points, the feature vector based on all main broadcasters generates main broadcaster and recommends mould Type, main broadcaster's recommended models are for selecting user main broadcaster to be recommended.

Live content recommended method provided by the present application, device and system select main broadcaster to be recommended for user；Generate institute Main broadcaster's recommendation list of user is stated, main broadcaster's recommendation list is used to record the relevant information of the main broadcaster to be recommended；To sight Many clients send main broadcaster's recommendation list.Present applicant proposes the new live streaming way of recommendation, the angle recommended from main broadcaster is User selects main broadcaster to be recommended, is watched with the related live video for facilitating user to select oneself interested main broadcaster.

Certainly, any product for implementing the application does not necessarily require achieving all the advantages described above at the same time.

Detailed description of the invention

In order to more clearly explain the technical solutions in the embodiments of the present application, make required in being described below to embodiment Attached drawing is briefly described, it should be apparent that, the drawings in the following description are only some examples of the present application, for For those of ordinary skill in the art, without any creative labor, it can also be obtained according to these attached drawings His attached drawing.

Fig. 1 is the Sample Scenario figure of the application in practical applications；

Fig. 2 is a kind of structure chart that recommender system is broadcast live provided by the embodiments of the present application；

Fig. 3 is a kind of flow chart of live streaming recommended method of service end side provided by the embodiments of the present application；

Fig. 4 is the flow chart of main broadcaster's recommended models training method provided by the embodiments of the present application；

Fig. 5 is the exemplary diagram of 68 critical point detections of face provided by the embodiments of the present application；

Fig. 6 is a kind of flow chart of live streaming recommended method of spectator client side provided by the embodiments of the present application；

Fig. 7 is a kind of schematic diagram that spectator client provided by the embodiments of the present application shows main broadcaster's recommendation list；；

Fig. 8 is another schematic diagram that spectator client provided by the embodiments of the present application shows main broadcaster's recommendation list；

Fig. 9 is another schematic diagram that spectator client provided by the embodiments of the present application shows main broadcaster's recommendation list；

Figure 10 is another schematic diagram that spectator client provided by the embodiments of the present application shows main broadcaster's recommendation list；

Figure 11 is the flow chart of another live streaming recommended method of service end side provided by the embodiments of the present application；

Figure 12 is the generation method flow chart of commodity classification recommendation list provided by the embodiments of the present application；

Figure 13 is a kind of flow chart of live streaming recommended method of main broadcaster's client-side provided by the embodiments of the present application；

Figure 14 is the schematic diagram that main broadcaster's client provided by the embodiments of the present application shows the commodity classification recommended；

Figure 15 is another schematic diagram that main broadcaster's client provided by the embodiments of the present application shows the commodity classification recommended；

Figure 16 is a kind of structure chart of live streaming recommendation apparatus of service end side provided by the embodiments of the present application；

Figure 17 is a kind of structure chart of live streaming recommendation apparatus of spectator client side provided by the embodiments of the present application；

Figure 18 is a kind of structure chart of live streaming recommendation apparatus of main broadcaster's client-side provided by the embodiments of the present application；

Figure 19 is a kind of structure chart of main broadcaster's recommended models training device provided by the embodiments of the present application.

Specific embodiment

Below in conjunction with the attached drawing in the embodiment of the present application, technical solutions in the embodiments of the present application carries out clear, complete Site preparation description, it is clear that described embodiments are only a part of embodiments of the present application, instead of all the embodiments.It is based on Embodiment in the application, it is obtained by those of ordinary skill in the art without making creative efforts every other Embodiment shall fall in the protection scope of this application.

Technical solution provided by the present application in order to facilitate understanding below first carries out the research background of technical scheme Simple declaration.

With the fast development of live streaming, more and more people select to obtain money by live streaming platform viewing live streaming News, live streaming bring many conveniences to daily life and work.Live data amount is skyrocketed through on live streaming platform, How to give user recommended user possible interested resource from a large amount of live streaming resource, this is the live streaming current urgent need to resolve of platform The technical issues of.

For platform is broadcast live, how within the shortest time that user stops to user to recommend its interested resource, Improve resource recommendation accuracy, this be influence user experience a key function point, be also influence user watch frequency and One key factor of duration, this characteristic also extreme influence live streaming platform development.

As live streaming platform, it is necessary to solved with more scientific method how within the shortest time that user stops to user Recommend its most interested live streaming relevant information, to improve user experience, increases the viewing frequency and duration of user.

Inventor's discovery: since live streaming is by the interactive mode of transmission of video information, user relatively takes notice of vision Experience, more emphasis Visual Aesthetics, user prefer to the related live streaming view for the main broadcaster for seeing that eye impressions meet oneself aesthetical standard The recommendation of frequency.Based on this, present applicant proposes a kind of live streaming recommended method, device and system, the angle recommended with main broadcaster is proposed Degree, excavating main broadcaster becomes a special characteristic attribute of live video content to the eye impressions that spectators generate, and being based on should Characteristic attribute is intelligently that different user recommends the main broadcaster's relevant information for meeting its practical aesthetic preference, to facilitate user according to this Main broadcaster's relevant information selects the live video of oneself interested main broadcaster to watch, to meet the personalization of different user Demand.

It should be noted that technical solution provided by the present application can be adapted for any live streaming platform, and such as: it is suitable for trip Class of playing live streaming platform, electric business class live streaming platform, amusement class live streaming platform, etc., for improving the resource recommendation that platform is broadcast live Accuracy improves user experience.

First the application scenarios of the application in practice are introduced below.

It is the Sample Scenario figure of the application in practical applications, as shown in Figure 1, technical side provided by the present application referring to Fig. 1 Case is applied in server-side 101, spectator client 102 and main broadcaster's client 103；Server-side 101, which refers to, provides live data service Equipment；Spectator client 102 refer to spectators watch live streaming when it is used, support video playing terminal device, as mobile phone, Laptop etc.；Main broadcaster's client 103 refers to that main broadcaster is used when publication is broadcast live, has video and audio recording and video The terminal device of transfer function.

In practical applications, server-side 101 can provide data support simultaneously for multiple spectator clients 102, can also be simultaneously It provides data for multiple main broadcaster's clients 103 to support, server-side 101 can be independent server in actual deployment, can also To be realized using cluster server.

For the application when realizing, server-side 101 learns obtained main broadcaster's recommended models, is to use by main broadcaster's recommended models The user of spectator client 102 generates corresponding main broadcaster's recommendation list, and is sent to spectator client 102, which passes through sight The live streaming relevant information that many clients 102 are shown, selects oneself interested live video to be watched.In addition, server-side 101 Commodity classification recommendation list can also be sent to main broadcaster's client 103, so that main broadcaster's client 103 is recommended according to the commodity classification List is that main broadcaster shows dependent merchandise classification, selects some commodity classification as live streaming material for main broadcaster.

Based on use above scene, this application provides a kind of live streaming recommender systems, and the system is introduced below.

It referring to fig. 2, is a kind of structure chart that recommender system is broadcast live provided by the embodiments of the present application, as shown in Fig. 2, the system Include:

Server-side 201 generates main broadcaster's recommendation list of the user for selecting main broadcaster to be recommended for user, described Main broadcaster's recommendation list is used to record the relevant information of the main broadcaster to be recommended, sends the main broadcaster to spectator client and recommends column Table；Preferably, the treatment process of server-side 201 can be found in the realization of embodiment illustrated in fig. 3.

Spectator client 202, the main broadcaster's recommendation list sent for receiving the server-side, pushes away according to the main broadcaster Recommending list is that user shows live streaming relevant information；Preferably, the treatment process of server-side 202 can be found in embodiment illustrated in fig. 6 It realizes.

Optionally, which in specific implementation, can also include:

Main broadcaster's client 203 is that main broadcaster generates commodity classification recommendation list for receiving the server-side 201；The quotient Category mesh recommendation list record pays close attention to the end article classification of the user group of the main broadcaster；The end article class of the user group Mesh is to be obtained according to the user group to the historical operation behavioural analysis of commodity.Preferably, the processing of main broadcaster's client 203 Process can be found in the realization of embodiment illustrated in fig. 13.

When specific implementation, server-side 201 can be applied in server terminal shown in FIG. 1 realize, to realize its function； And spectator client 202 can be applied with realizing in spectators' customer equipment shown in FIG. 1, to realize its function；Main broadcaster's client 203 can be applied in main broadcaster's customer equipment shown in FIG. 1 realize, to realize its function.

Using live streaming recommender system provided by the embodiments of the present application, main broadcaster to be recommended is selected for user；Generate the use Main broadcaster's recommendation list at family, main broadcaster's recommendation list are used to record the relevant information of the main broadcaster to be recommended；To spectators visitor Family end sends main broadcaster's recommendation list.Present applicant proposes the new live streaming way of recommendation, the angle recommended from main broadcaster is user Main broadcaster to be recommended is selected, is watched with the related live video for facilitating user to select oneself interested main broadcaster.

Live streaming recommended method provided by the present application is introduced below.

It is the flow chart of the live streaming recommended method of service end side provided by the embodiments of the present application referring to Fig. 3, as shown in figure 3, This method comprises:

Step 301: main broadcaster to be recommended is selected for user.

In the embodiment of the present application, the main broadcaster that server-side has focused on according to user is that user selects main broadcaster to be recommended, It can be that user selects main broadcaster to be recommended according to the main broadcaster for having the other users of network interaction of interest with user.Due to user Relatively take notice of visual experience when watching live streaming, more emphasis Visual Aesthetics, user can preferentially select appearance, stature or other Characteristics of human body meets oneself aesthetic main broadcaster, can select the live content for watching these main broadcasters.Therefore, the embodiment of the present application provides A kind of optional mode, using main broadcaster to user eye impressions as the characteristic attribute of main broadcaster, meet as user's selection The main broadcaster of user's Visual Aesthetics standard.

A kind of optional mode, according to the characteristics of human body of all main broadcasters in live data library be user select it is to be recommended Main broadcaster.Further, in order to improve the validity for recommending efficiency and recommendation, the embodiment of the present application, which also proposed, utilizes model It is handled, specifically, being that user selects main broadcaster to be recommended using main broadcaster's recommended models, wherein main broadcaster's recommended models are The mathematical model that the characteristics of human body of main broadcaster in live data library is obtained as training data training.Optionally, which pushes away Recommending model is to carry out clustering generation by feature vector of the clustering algorithm to each main broadcaster in live data library, In, described eigenvector is the multi-C vector for characterizing human face feature.

Wherein, the characteristics of human body of main broadcaster refers to from the beginning main broadcaster's human body whole feature or local feature, such as entire people arrive The characteristics of human body of foot, alternatively, face feature, hand-characteristic, step feature, leg feature etc., but in view of common at present big Crowd generally first navigates to face to the aesthetic of people, and therefore, optionally, the characteristics of human body of the main broadcaster is face in the embodiment of the present application Portion's feature, but the embodiment of the present application does not limit other forms.

Recommend in order to enable server-side provides personalized live streaming using live streaming recommended models for user, the embodiment of the present application Additionally provide a kind of main broadcaster's recommended models training method.Main broadcaster's recommended models training method is explained below.

Referring to fig. 4, Fig. 4 is a kind of flow chart of main broadcaster's recommended models training method provided by the embodiments of the present application, the party Method can be acquired the image comprising characteristics of human body of a large amount of main broadcaster when realizing by server-side, be carried out under line based on these images Processing training obtains main broadcaster's recommended models, and the training process of server-side training main broadcaster's recommended models is process under a line, this Process can be implemented regularly (such as on a daily or weekly basis), offline with server-side, and the result of this process can seen every time Crowd is called by server-side in real time when logging in the live streaming platform, as shown in figure 4, method includes the following steps:

401, obtain the image comprising characteristics of human body of main broadcaster；

In the embodiment of the present application, whole characteristics of human body of server-side available main broadcaster from live data library, or Partial body's feature, such as the head-to-toe characteristics of human body of entire people, alternatively, face feature, hand-characteristic, step feature, leg Portion's feature etc., but in view of current ordinary populace generally first navigates to face to the aesthetic of people, therefore, optionally, in this Shen It please the characteristics of human body of the main broadcaster be face feature in embodiment, but the embodiment of the present application does not limit other forms.

Optionally, described image is the face image comprising face feature.Below only by taking the face feature to main broadcaster as an example Training obtains main broadcaster's recommended models, but it's not limited to that for the training of main broadcaster's recommended models of the embodiment of the present application, to other portions The handling principle of the characteristics of human body divided is same or similar with the handling principle to face feature, no longer illustrates one by one herein It is bright.

In specific implementation, it when main broadcaster is broadcast live using live streaming platform publication for the first time, needs to register on live streaming platform personal Account needs to input account-related information when registering personal account, such as main broadcaster's Real Name, ID card No., cell-phone number Code, mailbox, account name, login password, the photo comprising characteristics of human body, such as personal whole body shine, and face shines, and foot shines etc.；It is then straight These information inputted when main broadcaster's login account can be saved by broadcasting platform.After main broadcaster completes registration, when needing to issue live streaming, main broadcaster Personal account is logged in using the account name and login password of registration, to issue live streaming.

Live streaming platform stores the account-related information of each main broadcaster, in addition, live streaming platform is also after main broadcaster issues and is broadcast live The live video that the main broadcaster is issued can be stored, for spectators' viewing.In specific implementation, the above information is all storable in Server-side.

In specific implementation, server-side can extract the personal full face of main broadcaster from the account-related information of main broadcaster, Using the photo as the facial image of main broadcaster.Server-side can also pass through face recognition algorithms from the live video resource of main broadcaster Interception obtains the facial image of main broadcaster.In practical applications, server-side can according to image processing algorithm to the size of image, as The requirement such as element, is adaptively adjusted the facial image got, to meet image processing requirements.

402, detection described image obtains the key point of characterization organization of human body feature；

In specific implementation, num can be detected for the facial image of each main broadcaster according to face critical point detection algorithm A key point；This num key point can portray the face mask of main broadcaster and the face location point of face position；

In specific implementation, server-side can use based on facial eyes structure feature, be based on SDM algorithm, based at random Forest algorithm realizes above-mentioned function based on any one face critical point detection algorithm of deep neural network algorithm etc.；It should Face critical point detection algorithm is for that can characterize the key point of face mask and face position, including face wheel on locating human face Exterior feature, eyes, eyebrow, lip and nose profile return to the key point coordinate position of human face five-sense-organ and profile.Algorithm detection Keypoint quantity and position embody the precision of the facial contour and face portrayed using the algorithm, portray detection with high accuracy and calculate The face key point of method can perfect fitting face.Therefore, the keypoint quantity num of detection can be embodied for detection accuracy It is required that in specific implementation, server-side can preset crucial points according to the required precision for facial contour and face Amount.Preferably, server-side presets num=68.

For example: using face critical point detection algorithm is based on, the face image of main broadcaster A is detected to obtain 68 Key point, the distribution of this 68 key points is as shown in figure 5, can accurately find out the face of the main broadcaster A according to this 68 key points Contouring and face position.

403, the corresponding feature vector of each main broadcaster is calculated according to the corresponding key point of each image；

Wherein, feature vector be calculated according to the key point of facial image, main broadcaster's face mask can be characterized and The vector of five features.Server-side can pre-define the calculating side of the dimension of this feature vector and the characteristic value of every dimension Formula, the face feature that the characteristic value of every dimension is characterized are all different；Such as it is arranged between the characteristic value eye of the first dimension Away from, the characteristic value nose length of the second dimension, characteristic value lip width of third dimension, etc..For example: still Based on the face image of main broadcaster A shown in fig. 5,15 critical lengths or crucial distance are calculated according to key point, it is fixed The calculation of the characteristic value of adopted 14 dimensional feature vectors and each dimension, obtains 14 dimensional feature vectors of main broadcaster A, specific to calculate Step are as follows:

It is as follows to obtain wherein pass key length or range data on the basis of these known 68 key points by S1:

1) length of nose, the i.e. midpoint Fig. 2 27 arrive the fore-and-aft distance of point 33, are denoted as a；

2) pupil is to the fore-and-aft distance of lip, i.e. the geometric center at the midpoint Fig. 2 37,38,40,41 and point 61,63,65,67 Geometric center fore-and-aft distance, be denoted as b；

3) width of nose, the i.e. midpoint Fig. 2 31 arrive the lateral distance of point 35, are denoted as c；

4) width of lip, the i.e. midpoint Fig. 2 48 arrive the lateral distance of point 54, are denoted as d；

5) spacing in nostril, the i.e. midpoint Fig. 2 32 arrive the lateral distance of point 34, are denoted as e；

6) geometric center of the spacing of pupil, the i.e. geometric center at the midpoint Fig. 2 37,38,40,41 and point 43,44,46,47 Lateral distance, be denoted as f；

7) the interior spacing of two eyebrows, the i.e. midpoint Fig. 2 21 arrive the lateral distance of point 22, are denoted as g；

8) left eye right eye angle is at a distance from right eye left eye angle, i.e., the midpoint Fig. 2 36 is denoted as h to the lateral distance for putting 45；

9) width of face transverse direction the widest part, the i.e. midpoint Fig. 20 arrive the lateral distance of point 16, are denoted as q；

10) width of single eyebrow, the i.e. midpoint Fig. 2 17 arrive the lateral distance of point 21, are denoted as j；

11) width of single eyes, the i.e. midpoint Fig. 2 36 arrive the lateral distance of point 39, are denoted as k；

12) opening degree of single eyes, the i.e. midpoint Fig. 2 37 arrive the linear distance of point 40, are denoted as l；

13) mouth is denoted as m to the fore-and-aft distance of chin, the i.e. fore-and-aft distance at the midpoint at the midpoint Fig. 2 62,66 and point 8；

14) pupil is to the fore-and-aft distance of nose, i.e., the geometric center at the midpoint Fig. 2 37,38,40,41 and point 33 it is longitudinal away from From being denoted as n；

15) nose is denoted as o to the fore-and-aft distance of mouth, the i.e. fore-and-aft distance at the midpoint Fig. 2 33 and point 51.

S2 defines the feature vector of 14 dimensions, is denoted as vector type variable X, for some specific value of X, is denoted as vector x. It is defined as follows to wherein often one-dimensional:

X [1]=b/a；

X [2]=d/c；

X [3]=c/e；

X [4]=f/g；

X [5]=h/q；

X [6]=j/q；

X [7]=d/q；

X [8]=k/q；

X [9]=k/l；

X [10]=h/n；

X [11]=n/o；

X [12]=m/o；

X [13] indicates whether to wear glasses, if having worn glasses, X [13]=1.0, and if do not worn glasses, X [13]= 0.0；

X [14] indicates gender, if it is male, then [14]=1.0 X, and if it is women, then [14]=0.0 X.

According to the definition of above-mentioned 14 dimensional feature vector calculation method, available characterization main broadcaster A face mask and face The feature vector of 14 dimensions.

There is M main broadcaster in live streaming platform, anchor_i (i=1,2 ... M, M indicate main broadcaster's total number of persons) is denoted as, according to above The corresponding feature vector of each main broadcaster is calculated in mode, characterizes the facial contour and face of each main broadcaster.

404, main broadcaster's recommended models are generated according to the feature vector of all main broadcasters.

In specific implementation, clustering is carried out to the corresponding feature vector of all main broadcasters using clustering algorithm to obtain Feature clustering matrix is denoted as main broadcaster's recommended models.

Clustering is carried out to the feature vector of all main broadcasters, the purpose of the clustering is to establish one to contain K group The distributed model of class.Wherein, K is denoted as the number of class, and expression divides main broadcaster according to the feature vector of all main broadcasters Total classification of class.

Clustering is carried out to the feature vector of main broadcaster, adoptable clustering algorithm includes: k-means clustering algorithm, height This mixed model (Gaussian Mixture Model, GMM) algorithm etc..It is only illustrated by taking GMM algorithm as an example below.By GMM algorithm obtains probability density function, can measure the probability that main broadcaster belongs to certain group class classification by probability density function Value.In specific implementation, step 404 can be divided into following steps:

Firstly, the distribution function of single class is as follows:

Wherein, x is the feature vector acquired in above-mentioned steps 403, and u is the internal model expectation of the class, and ∑ is this The internal model variance of class.

Then, according to the distribution function of single class defined above, for all main broadcasters cluster analysis result such as Under:

Wherein, K is class number, π_kIndicate the weight of k-th of class.In specific implementation, K is according to experience It is preset, can characterize the face feature classification number of people, such as can with value for 10,20,30, etc. integer values, in order to obtain Finer category division, K can be using values as bigger numerical, and K can be the integer between 10 to 100 with value.Certain the application Embodiment does not limit the specific value condition of K.

Then, by feature extraction above, M feature vector x is obtained_i(i=1,2 ... M, M indicate the total people of main broadcaster Number)；Using M obtained feature vector as M sample, pass through EM algorithm (Expectation Maximization Algorithm, EM algorithm), acquire the parameter u in above-mentioned expression formula (1) and expression formula (2)_k,∑_k,π_k。

Wherein, EM algorithm is that a kind of iteration in data clusters field for being frequently used in machine learning and computer vision is calculated Method carries out maximal possibility estimation to the probability parameter model containing hidden variable (latent variable) or maximum posteriori is general Rate estimation.The calculating process of EM algorithm are as follows: the first step calculates expectation (E), utilizes the existing estimated value of probabilistic model parameter, meter Calculate the expectation of hidden variable；Second step, maximize (M), the expectation of the hidden variable acquired is walked using E, to parameter model into Row maximal possibility estimation；The estimates of parameters found in M step is used for during next E step calculates, this process constantly alternately into Row.

Finally, for each main broadcaster anchor_i (i=1,2 ... M), by its feature vector x_iSubstitute into expression formula above (1) in, its ascribed value for corresponding to k-th of class is acquired, anchor_i_in_class_k (k=1,2 ... K) are denoted as.

It then obtains: anchor_i_in_class_k=N (x_k；u_k,∑_k)。

As a result, by above-mentioned training process, the feature clustering matrix of M*K is established, AnchorToClass, this feature are denoted as Clustering matrix is main broadcaster's recommended models.The element that this feature clusters the i-th row k column in matrix is anchor_i_in_class_ K, wherein i=1,2 ... M；K=1,2 ... K.

Specifically, anchor_i_in_class_k indicates that the face feature of i-th of main broadcaster belongs to the k of main broadcaster's recommended models The probability value of a classification, i.e., above-mentioned ascribed value.

For in M main broadcasters on earth homography M row which, can be according to its temperature on the live streaming platform Ranking determines, is also possible to be determined according to the morning and evening of registion time, or determined according to the live video quantity of each main broadcaster, etc. Deng.

For the corresponding relationship of specific line number and main broadcaster, can be it is corresponding with the login account name of main broadcaster, can also be with The live video link of main broadcaster is corresponding.In specific implementation, for how to determine the corresponding relationship gone in main broadcaster and matrix, how The detailed process of specific main broadcaster is corresponded to according to line number, the application is not construed as limiting.

For example: based on the still above citing and scene shown in fig. 5, in the live streaming platform where main broadcaster A, main broadcaster Number is 50, and according to above-mentioned treatment process, server-side training obtains the feature clustering matrix that main broadcaster's recommended models are a 50*14. It is number three according to temperature of the main broadcaster A on the live streaming platform, the main broadcaster for being 3 as number in this learning model, i.e., The main broadcaster of the third line, by the association corresponding with " 3 " this line number of the name on account " A " of main broadcaster A；The third line in main broadcaster's recommended models Each element, the ascribed value that as main broadcaster A belongs to k type.

Main broadcaster's recommended models can be pre-established by method shown in Fig. 4, then select main broadcaster to be recommended to accomplish fluently number for user According to basis.What needs to be explained here is that the server-side for carrying out the training of main broadcaster's recommended models selects main broadcaster to be recommended with for user Server-side can be same server-side, or independent different server-side.

Step 302 is continued to execute after server-side executes step 301 with continued reference to Fig. 3.302, generate the user's Main broadcaster's recommendation list, main broadcaster's recommendation list are used to record the relevant information of the main broadcaster to be recommended；

303, main broadcaster's recommendation list is sent to spectator client.

For server-side, when user logs in live streaming platform, step 301-303 is just executed in real time.

In practical applications, some users are uncomfortable implements live streaming relevant operation, such as collection, point on live streaming platform The operation such as praise, comment on, therefore, live streaming platform can not be collected into the historical behavior data of this kind of spectators；In addition, for stepping on for the first time For the user of Lu Pingtai, also without generating any historical behavior data.For this kind of users, tradition live streaming platform is to user It is recommended that the live streaming resource that live streaming click volume is relatively high, resource type is not comprehensive enough, does not analyze user's factor interested, Possible interested live streaming resource is provided for user, causes this kind of users interested due to can not find oneself in a short time Resource is broadcast live, and quickly leaves live streaming platform, flow is caused to be lost, this has lost live streaming platform very big.Based on this Scape demand, the embodiment of the present application provide corresponding settling mode, which is to recommend mould according to main broadcaster trained in advance Main broadcaster is selected a variety of different types of main broadcasters for user, is based on this, used by type to the eye impressions of spectators as characteristic attribute After family logs in live streaming platform, oneself interested main broadcaster can be rapidly selected from a variety of different types of main broadcasters, helps to use Family quickly positions oneself point of interest to live streaming.

Based on above-mentioned main broadcaster's recommended models, the embodiment of the present application provides several optional implementations for step 301:

It is optionally, described to select main broadcaster to be recommended according to main broadcaster's recommended models for user, comprising:

According to main broadcaster's recommended models, the maximum main broadcaster of ascribed value in each main broadcaster's classification is selected, selected main broadcaster is made For main broadcaster to be recommended；Wherein, the ascribed value refers to that main broadcaster belongs to the probability value of some main broadcaster's classification.

According to main broadcaster's recommended models, the maximum main broadcaster of ascribed value in each main broadcaster's classification is selected；It is flat in live streaming according to main broadcaster The Flow Value generated on platform, the main broadcaster for the predetermined number for selecting Flow Value in the top from selected main broadcaster, will select The main broadcaster selected is as main broadcaster to be recommended.

Wherein, the Flow Value that main broadcaster generates on live streaming platform is regarded according to the live video quantity of main broadcaster's publication, live streaming Frequency clicking rate, the parameter any one or more such as commodity transaction conversion ratio of live video determine.Such as: the live video of main broadcaster's publication More, the Flow Value generated is bigger；The video click rate of main broadcaster's publication is higher, then the Flow Value of its generation is bigger, etc. Deng.Optionally, to be method that user selects main broadcaster to be recommended according to above-mentioned main broadcaster's recommended models, comprising:

According to main broadcaster's recommended models, the maximum main broadcaster of ascribed value in each main broadcaster's classification is selected, further according in live streaming platform Popular main broadcaster's information selects at least two popular main broadcaster's information, finally, by selected at least two from selected main broadcaster A hot topic main broadcaster is as main broadcaster to be recommended；Wherein, the ascribed value refers to that main broadcaster belongs to the probability value of some main broadcaster's classification.

It " according to main broadcaster's recommended models, selects to return in each main broadcaster's classification to involved in above several preferred embodiments below The specific implementation of this operation of the maximum main broadcaster of category value " is explained.

In specific implementation, for spectators client User_p, wherein p=1,2 ... N, N are live streaming platform spectators client Sum is searched the maximum element of ascribed value in each column and is denoted as according to the main broadcaster recommended models AnchorToClass pre-established Max_in_class_k, k=1,2 ... K；And line number corresponding to this K maximum ascribed value is recorded, max_anchor_ is denoted as In_class_k, k=1,2 ... K.Each line number uniquely corresponds to a main broadcaster, then is capable of determining that specific main broadcaster according to line number, K main broadcaster is then determined, using this K main broadcaster as main broadcaster to be recommended.Since this K main broadcasters to be recommended are that can represent respectively The typical main broadcaster of this K class main broadcaster's type classified based on face feature, this K main broadcasters to be recommended represent the aesthetic type of K kind, With different face masks and five features, determine that user can be quick to the main broadcaster to be recommended of user based on this K main broadcaster Make a decision oneself interested main broadcaster out and related live video.

By the above method, the embodiment of the present application can provide a variety of different selections from Visual Aesthetics angle for user, Facilitate user after entering live streaming platform, is quickly determined for compliance in the short time in oneself aesthetic main broadcaster and live video Hold.

In practical applications, some users have good live streaming platform interaction habits, they can make when watching live streaming It the operation such as collects, thumb up, commenting on, this kind of users their respective hobbies can be analyzed from these operation behaviors and (liked Main broadcaster, the live video type liked).Therefore, for a kind of user, platform is broadcast live and needs more accurately to recommend live streaming phase to it Information is closed, invalid recommendation is avoided result in, this kind of users is interfered, user experience is influenced.Based on this application scenarios, this Shen Please embodiment provide corresponding solution, the solution is according to the masters of the historical operation behavioural analysis user preferences of spectators The face feature broadcast, to recommend the relevant information for other main broadcasters for meeting its aesthetical standard for user, on the one hand guarantee On the other hand the accuracy of recommendation helps user to extend its point of interest.The specific method is as follows:

For the above-mentioned spectators for having the operation relevant to specific video program such as viewing, collection, the embodiment of the present application is also mentioned The following method for selecting main broadcaster to be recommended according to main broadcaster's recommended models for user is supplied.

Optionally, the method for selecting main broadcaster to be recommended according to above-mentioned main broadcaster's recommended models for user, comprising:

Historical operation behavior according to user to live streaming determines user actually interested main broadcaster, is denoted as first kind main broadcaster；

According to main broadcaster's recommended models, the similarity distance between other main broadcasters and the first kind main broadcaster is calculated；It is described Other main broadcasters refer to the main broadcaster removed except the first kind main broadcaster in all main broadcasters；

According to the size relation of similarity distance, the main broadcaster of predetermined number is selected from other described main broadcasters, it will be selected Main broadcaster is as main broadcaster to be recommended.

In specific implementation, historical operation behavior refers to operation relevant to the live streaming row that user implements on live streaming platform For if user is in the live video search behavior of search box, live video checks behavior, thumbs up behavior, right to live video The collection behavior of live video, to the collection behavior of main broadcaster, to comment behavior of live video, etc..

In specific implementation, determine that user is actually interested in the historical operation behavior of the live streaming platform according to spectators Main broadcaster is denoted as R main broadcaster, and R is the integer numerical value more than or equal to 1, this R main broadcaster has unique encodings in live streaming platform, Be denoted as respectively interest_anchor_1, interest_anchor_2 ... interest_anchor_R, by this determining R Main broadcaster is denoted as first kind main broadcaster.

Wherein, in specific implementation, server-side can determine the number R, Yi Zhongfang of first kind main broadcaster in the following manner Formula is that R value is HistoryCnt, and HistoryCnt is that the user that server-side goes out according to the historical operation behavioural analysis of spectators is real The total number of the interested all main broadcasters in border；Another way is, server-side according to value formula min (R1, HistoryCnt), I.e. R value is the minimum value in R1 and HistoryCNT, and wherein R1 is that platform set main broadcaster according to actual needs is broadcast live Number, such as R1 value are 3, and in practical applications, in order to reduce operand, R1 value is the units greater than 1.

After determining first kind main broadcaster, server-side utilizes main broadcaster's recommended models, calculates other main broadcasters and this R main broadcaster Between similarity distance, wherein other main broadcasters refer to that all main broadcasters remove main broadcaster except this R main broadcaster in live streaming platform.Example Such as: for main broadcaster anchor_i (1≤i≤M), be not belonging to main broadcaster interest_anchor_1, interest_anchor_2 ... Interest_anchor_R then calculates the similarity distance of main broadcaster anchor_i Yu this R main broadcaster, and specific formula for calculation is as follows:

According to above-mentioned formula (3), the similarity distance dist_i of other main broadcasters Yu this R main broadcaster are calculated, then, according to phase Like the size relation of distance, the main broadcaster of predetermined number is selected from other described main broadcasters, using selected main broadcaster as to be recommended Main broadcaster.For example, predetermined number is h, then the lesser h main broadcaster of similarity distance is selected from other main broadcasters, this h main broadcaster is made For main broadcaster to be recommended.Since similarity distance is smaller, illustrates that the face feature of two main broadcasters is more similar, then according to similarity distance For the main broadcaster to be recommended of user's selection, meet the practical aesthetical standard of user.

Using aforesaid way, the embodiment of the present application can be provided more according to the historical operation behavior of spectators for spectators Meet the relevant information of the main broadcaster to be recommended of its practical aesthetical standard, provides more selections with drawings family, promote user experience.

It is the introduction of the live streaming recommended method embodiment provided by the present application in server-side above, will be stood below in spectators visitor Family end is described live streaming recommended method.

It is the flow chart of the live streaming recommended method of spectator client side provided by the embodiments of the present application, such as Fig. 6 referring to Fig. 6 It is shown, this method comprises:

Step 601, main broadcaster's recommendation list that server-side is sent is received, main broadcaster's recommendation list is to be recommended for recording The relevant information of main broadcaster；

Main broadcaster's number to be recommended can be preset using the spectators client of live streaming platform, is sent receiving server-side Main broadcaster's recommendation list when, pre-set main broadcaster to be recommended several main broadcasters are first selected, and by its relevant information, as will It is that user recommends valuable information to make data preparation that the content to be recommended, which is subsequent,.

It step 602, is that user shows live streaming relevant information according to main broadcaster's recommendation list.

Optionally, step 602 can be realized as follows:

It is that user shows main broadcaster's information to be recommended, main broadcaster's information to be recommended according to main broadcaster's recommendation list, comprising: The network connection for the live video that main broadcaster's title, the face image of main broadcaster, the live video of main broadcaster's publication, main broadcaster issue, waits one Kind or much information.

In specific implementation, spectator client recommends page to show main broadcaster's title, spectators according to main broadcaster's recommendation list in main broadcaster These main broadcaster's titles (such as: some stars, net are red) is seen, by oneself previous eye impressions to these main broadcasters, to determine Oneself interested main broadcaster, and then the related live video of oneself interested main broadcaster is searched in the search box of live streaming platform.

In specific implementation, spectator client recommends page to show main broadcaster's title, main broadcaster according to main broadcaster's recommendation list in main broadcaster Face image, related to the main broadcaster live video page data association of the face image of the main broadcaster；Then spectators see that main broadcaster recommends page The face image of displaying can quickly determine oneself interested main broadcaster, by clicking the face image of the main broadcaster, into the main broadcaster Live video page, select interested live video to be checked.

In specific implementation, spectator client recommends page to show: main broadcaster's title, master according to main broadcaster's recommendation list in main broadcaster The face image broadcast and the live video of main broadcaster publication；Then spectators see the face image that main broadcaster recommends page to show, can be fast Speed determines oneself interested main broadcaster, and then recommends page to click directly in the main broadcaster and check the related live video of the main broadcaster.

It optionally, is that user shows live streaming relevant information according to main broadcaster's recommendation list, comprising:

The relevant information of the main broadcaster of predetermined number is selected from main broadcaster's recommendation list；

Recommend the relevant information of the main broadcaster of page presentation selection in live streaming, the relevant information of the main broadcaster includes: main broadcaster's The network linking of face image live video related to main broadcaster.

In specific implementation, live streaming platform or the spectators client using the live streaming platform can preset to be recommended Main broadcaster's number, according to the main broadcaster's recommendation list for receiving server-side transmission, therefrom selection meets the master for presetting number requirement It broadcasts, as the main broadcaster that will recommend, and is associated with its relevant information.Wherein, the relevant information of main broadcaster includes: face's figure of main broadcaster The network linking of picture, main broadcaster's correlation live video, also may include other information such as main broadcaster's personal information, height, weight, educational background Etc..

Then, spectator client shows the relevant information of above-mentioned main broadcaster to be recommended, and spectators client can scheme according to face As the main broadcaster for selecting to meet oneself aesthetical standard, its face image is clicked, all live videos of the main broadcaster can be presented, in institute Have and oneself possible interested live streaming is selected to watch in the live video of the main broadcaster, specific effect is as shown in Figure 7.In Fig. 7 In, left hand view is shown main broadcaster and recommends the page, shows three main broadcasters on the page, the face feature of these three main broadcasters not phase Together, recommend to be provided with corresponding display area on the page in the main broadcaster for each main broadcaster, which is divided into two pieces up and down Region is used to show the live streaming theme of the main broadcaster above for showing the face image of main broadcaster below, such as can be the main broadcaster The theme of the highest live video of the click volume of publication or the customized live streaming theme of main broadcaster etc., the master shown due to one page Limited amount is broadcast, therefore, when the relevant information of all main broadcasters can not then be realized by page turning control when one page shows and finishes Multipage shows that there are many forms of page turning control, a kind of Fig. 7 only example.

Spectators like according to itself, and the main broadcaster for selecting oneself to like clicks the net of the live video below its face image Network link, the page jump, and recommend page to jump to live video from live streaming and recommend page, recommend page to show this in the live video The relevant live video of main broadcaster, to facilitate user that oneself interested live video is selected to check.In live video page Can sort each live video according to the click volume sequence of live video, can also sort according to issuing time sequence each A live video.

Spectators client can select the main broadcaster for meeting oneself aesthetical standard according to face image, and main broadcaster correlation is checked in click The network linking of live video, live streaming topic links as shown in Figure 8, the page of spectator client control at this time jump, from Main broadcaster recommends page to jump to live video broadcast page, then shows in the live video broadcast page and the associated tool of topic links is broadcast live Body live video,

As shown in figure 8, spectators click the live streaming topic links below the main broadcaster that some may like oneself, the page is jumped Turn, jumps and show specific live video on rear live video broadcast page, certainly, in specific implementation, can also be regarded in the live streaming Other live videos relevant to the main broadcaster are shown on frequency broadcast page, and user is facilitated to select.

By the above-mentioned method for showing main broadcaster's recommendation list for spectator client, it can allow spectators client intuitively from face Aesthetic angle selects face mask and face to meet oneself aesthetic main broadcaster, watches the video of its live streaming, improves live streaming and recommends Efficiency.

According to main broadcaster's relevant information in the live streaming recommendation list, the relevant live streaming of main broadcaster is obtained from live data library Video；

Recommend the live video that acquisition is shown on the page in live streaming.

In specific implementation, this method is found in the database of live streaming platform with the main broadcaster in live streaming recommendation list Live video relevant to above-mentioned each main broadcaster, and arranging from high to low by the end of current click volume according to live video Column either thumb up perhaps number of reviews according to the spectators of live video and arrange displaying main broadcaster's account name or number from high to low Etc. the unique identifications main broadcasters information, and make a reservation for several live videos in the corresponding region of the main broadcaster of each recommendation, and The live streaming of a certain live streaming platform of spectators' customer terminal equipment is recommended to show on the page.

Spectators client recommends to meet oneself aesthetical standard in the static display image for selecting live video on the page in live streaming Main broadcaster, if it is desired to select interested content to watch from all live videos of the main broadcaster, click its main broadcaster volume Number or account name, so that it may be linked to the page that all live videos of the main broadcaster are shown, select interested content viewing i.e. Can, specific effect is as shown in Figure 9.As shown in Fig. 9 left hand view, spectators click oneself interested main broadcaster's number or account name Claim, the page will jump to the display page of Fig. 9 right part of flg, show the face image and its base of the main broadcaster on the top of the page This information will be shown below all live videos of the main broadcaster, and user is facilitated to select oneself interested live video.

Spectators client recommends to meet oneself aesthetical standard in the static display image for selecting live video on the page in live streaming Main broadcaster, if recommend the page on live video have spectators client want viewing content, click directly on live streaming recommend the page On the live video, similarly, in the lower section or the right of the video of viewing, it may appear that the main broadcaster other live streaming view Frequently, alternately and subsequent recommendation, specific effect are as shown in Figure 10.

By the above-mentioned method for showing main broadcaster's recommendation list for spectator client, spectators client can be allowed directly to pass through recommendation The predetermined number live video of each main broadcaster understands while selecting face mask and face to meet oneself aesthetic main broadcaster The substantially live streaming type of the main broadcaster further accurately finds the live video for meeting oneself demand, improves live streaming and recommends efficiency.

In addition to it is above two spectators live streaming platform recommend the page on main broadcaster's display form, can also include certainly, Show that the form of main broadcaster's recommendation list is not only limited in above-mentioned several ways, in the embodiment of the present application to main broadcaster's recommendation list Exhibition method do not make considered critical.

The live streaming recommended method of spectator client side provided by the embodiments of the present application, main face image by main broadcaster or The live video still image of person main broadcaster is presented to spectators client's face mask and the different types of main broadcaster of face and its live streaming view Frequently；Or by, in the historical record of live streaming platform, analyzing, calculating the face for the main broadcaster that the spectators client likes to spectators client Feature, find out with the most similar main broadcaster of the main broadcaster, the video display being broadcast live is to spectators client.Such live streaming is recommended Based on spectators client for the aesthetical standard of main broadcaster's face feature, live streaming is made to recommend that more there is specific aim, and for live streaming For the spectators client of platform, can quickly, easily find the live video liked.

Optionally, other than the live streaming recommended method embodiment above in server-side, the application's the embodiment of the present application exists The live streaming recommended method of server-side further include recommend to main broadcaster live streaming may welcome live content, specific implementation below into Row description.

It is the flow chart of the live streaming recommended method of service end side provided by the embodiments of the present application, such as Figure 11 institute referring to Figure 11 Show, this method comprises:

Step 1101, commodity classification recommendation list is generated for main broadcaster；

The commodity classification recommendation list is used to record the end article classification for the user group for paying close attention to the main broadcaster；It is described The end article classification of user group is to be obtained according to the user group to the historical operation behavioural analysis of commodity.

In specific implementation, server-side is directed to some main broadcaster, obtains the historical operation behavior for paying close attention to the user of the main broadcaster, root According to the historical operation behavior of each user, the interested commodity classification of each user is analyzed, the main broadcaster is then directed to, counts its use Commodity classification interested to the group of family, commodity classification interested to user group is exactly end article classification, according to these End article type generates the corresponding commodity classification recommendation list of the main broadcaster, in case it is subsequent to each main broadcaster's client push, refer to It leads main broadcaster and selects the content to be broadcast live.

It should be noted that in the embodiment of the present application, the user for paying close attention to some main broadcaster to be also referred to as to the powder of the main broadcaster Silk, and the bean vermicelli of the main broadcaster refers to the main broadcaster had the user directly or indirectly interacted in live streaming platform；For example, being broadcast live The main broadcaster, the live video for checking main broadcaster publication, the live streaming collected the main broadcaster, collected the main broadcaster were searched in platform Video, etc., the user with above-mentioned behavior, the referred to as bean vermicelli of the main broadcaster.

Step 1102, the commodity classification recommendation list is sent to main broadcaster's client.

In specific implementation, server-side is generated according to the historical operation behavior of the user group of concern main broadcaster for the main broadcaster Corresponding commodity classification recommendation list, is sent to main broadcaster's client, to show that the commodity classification recommends column in main broadcaster's client The content of table, the theme and commodity classification being broadcast live next time for main broadcaster with reference to selection.

Optionally, commodity classification recommendation list is generated for main broadcaster, comprising:

Step 1103, for each user in the user group for paying close attention to the main broadcaster, it is of interest to obtain each user Merchandise news carries out the classification of commodity classification to the merchandise news；

In specific implementation, commodity classification recommendation list is generated for a certain main broadcaster, firstly, obtaining concern, collecting the master It broadcasts the perhaps user of its live video or watches the user of the live video of the main broadcaster；And by the institute of above-mentioned each user The merchandise news of concern carries out the classification of commodity classification.For example, commodity classification is divided into: clothes, shoes packet, electronics, food etc., and will The historical behavior of the interested user of each classification is counted.

Step 1104, according to each user to the historical behavior counting users of commodity to the interest value of commodity, according to commodity Interest value calculate the interest value of each commodity classification in the commodity classification；

In specific implementation, by the classification to commodity classification, according to each user to the historical behavior of commodity (as bought Whether quantity, browsing time, collection etc.) it is counted, and the interest value of every kind of historical behavior is distributed, then made with interest value For weight, the historical behavior number of number of users, commodity that every commodity of each commodity classification are related to and interest value into Row weighting, using obtained weighted value as the interest value of each commodity classification.

Step 1105, according to the interest value sequence size relation of commodity classification, commodity classification recommendation list is generated.

In specific implementation, according to the size of the interest value of each commodity classification, the sense of the user group of the main broadcaster is emerging Interesting commodity carry out a sequence, according to commodity classification interest value from big to small be sequentially generated commodity classification recommendation list, with The list is occurred main broadcaster is facilitated to choose the theme to be broadcast live and content to corresponding main broadcaster's client for subsequent.

Above-described embodiment, to behaviors such as collection, the purchases of commodity, recommends most probable to main broadcaster according to the bean vermicelli of recent main broadcaster Allow the interested merchandising classification of its fan group body.This process is related to the commodity preference of all beans vermicelli of the main broadcaster Statistics and analysis it is therefore proposed that regularly (such as daily), executes this process offline, and set time point will analysis knot Fruit feeds back to main broadcaster, for example, can timing 00 divide analysis result to each main broadcaster's feedback system at every night 24.

The embodiment of the present application additionally provides a kind of concrete implementation mode for how to generate commodity classification recommendation list, Referring to Figure 12, which includes:

S120 obtains the items list in the bean vermicelli collection, is denoted as List0 to each bean vermicelli；Obtain bean vermicelli purchase Items list in object vehicle, is denoted as List1；The items list in nearly 3 months history purchaser records of the bean vermicelli is obtained, is denoted as List2。

In specific implementation, usually there is a large amount of user to pay close attention to same main broadcaster, i.e. the main broadcaster possesses a large amount of bean vermicelli, in order to The commodity classification for accurately recommending its bean vermicelli group preferred for main broadcaster, it is necessary to each bean vermicelli in bean vermicelli group to commodity The preference profile of classification is analyzed, then needs to handle each bean vermicelli all in accordance with mode shown in Figure 12.

Since user adds products to collection, shopping cart or bought some commodity, these features are enough Illustrate user to the interest level of commodity, therefore, in the mode shown in Figure 12, using collection, shopping cart, purchaser record come Embody the interested commodity of each bean vermicelli and interest level.But it in specific implementation, can also be using bean vermicelli on platform Historical viewings record, search record etc. information it is interested in which commodity to analyze user and emerging to the specifically sense of commodity Interesting degree.

Total classification number scale of commodity is TypeNum by S121, and classification number is denoted as type_i (0≤i≤TypeNum), The statistics score value of classification type_i is denoted as intrestValue [i] (0≤i≤TypeNum)；Initialize intrestValue [i]=0.

The weight factor of collection is denoted as value0, all commodity in List0 is scanned, if some commodity belongs to by S122 Classification type_i then enables intrestValue [i]=intrestValue [i]+value0.

The weight factor of shopping cart is denoted as value1, all commodity in List1 is scanned, if some commodity belongs to by S123 Classification type_i then enables intrestValue [i]=intrestValue [i]+value1.

The weight factor of history purchaser record is denoted as value2, all commodity in List2 is scanned, if some quotient by S124 Product belong to classification type_i, intrestValue [i]=intrestValue [i]+value2.

Above-mentioned weight factor value0, value1, value2 can specifically be set according to the actual situation, not be It is changeless.In specific implementation, can be value0 >=value1 >=value2 or value0≤value1≤ value2.Three weight factors value0, value1, value2 can be both configured to identical value, also can be set wherein two A value is identical, may be set to be three different values, is not especially limited herein.

S125, to the score value intrestValue [i] (0≤i≤TypeNum) of each classification by sequence from big to small into Row sequence feeds back commodity category list of the ranking at first several to main broadcaster；For example, intrestValue [i] maximum preceding 10 A commodity classification.

By the method for the embodiment of the present application, server-side recommends the interested commodity classification of bean vermicelli to main broadcaster, facilitates main broadcaster Live streaming theme and specific Recommendations are determined according to the commodity classification of recommendation, and effect is broadcast live to improve it.

It is provided by the present application above in server-side, the live streaming recommended method embodiment of commodity classification push is carried out for main broadcaster Introduction, below will from the angle of main broadcaster's client to live streaming recommended method be introduced.

Referring to Figure 13, it is the flow chart of the live streaming recommended method of main broadcaster's client-side provided by the embodiments of the present application, such as schemes Shown in 13, this method comprises:

Step 1301, the commodity classification recommendation list that server-side is sent is received, the commodity classification recommendation list is for remembering The end article classification of the user group of record concern main broadcaster；The end article classification of the user group is according to the user group Body obtains the historical operation behavioural analysis of commodity；

In specific implementation, server-side sends the corresponding commodity classification recommendation column for belonging to each main broadcaster to main broadcaster's client Table, the list records about the end article classification of the user group of concern main broadcaster and to the value interested of each classification, And the commodity classification recommendation list sent should be that server-side is ranked up from big to small according to the value interested of commodity classification List afterwards.

It step 1302, is that main broadcaster shows commodity classification according to the commodity classification recommendation list.

In specific implementation, main broadcaster's client can specially be arranged a commodity classification and recommend the page, specially based on display Broadcast the commodity classification of recommendation；Main broadcaster's client can divide a display area on existing main broadcaster's personal homepage, aobvious at this Show the commodity classification that main broadcaster's recommendation is shown as in region；Main broadcaster's client can also be arranged one on existing main broadcaster's personal homepage A functionality controls, the functionality controls are used to check the commodity classification recommended for main broadcaster.

For example: the commodity classification that main broadcaster's client is shown, which is shown, referring to Figure 14, Figure 14 recommends schematic diagram,

In Figure 14, left hand view is main broadcaster's personal homepage, which includes for showing main broadcaster's relevant information Functional module, such as " personal essential information ", " my assets " (such as account balance, present, gold coin) ", " member's " center ", The functional modules such as " comment ", " collection "；One functionality controls " commodity classification recommendation list " is set on main broadcaster's personal homepage, By the functionality controls, facilitate main broadcaster to check for its recommend commodity classification.Main broadcaster can click the functionality controls, check that correlation pushes away Content is recommended, then page jump to specific commodity classification recommends the page, recommends to show commodity classification on page in the commodity classification.

Certainly, commodity classification recommends the display form of the page to be not limited to Figure 14, and the embodiment of the present application also proposed one Kind exhibition method, referring to Figure 15, in Figure 15, commodity classification is recommended to show the relevant information of each commodity classification on the page, such as The dependent merchandise of commodity class now, number interested and user are more straight in this way to interested accounting of each commodity etc. Seeing ground is main broadcaster's Recommendations classification, and main broadcaster is facilitated to determine live content next time according to the commodity classification.

The commodity classification that display in Figure 14 and Figure 15 is recommended is actual commodity classification, such as clothes, shoes packet, but It is that the commodity classification of actual recommendation is without being limited thereto, further includes the recommendation of the virtual goods such as some service classes, financial class, in this Shen Commodity classification and specific commodity please not be restricted in embodiment.

Recommended by the live streaming of the above-mentioned commodity classification recommendation list for receiving spectators may be welcome in main broadcaster's client The commodity classification of the user group for each main broadcaster for counting and being calculated is recommended column according to server-side by the embodiment of method Table, is sent to main broadcaster's client, and main broadcaster's client receives the list and shown and put down in the live streaming of oneself terminal device On the display page of platform, main broadcaster can intuitively with reference to and the theme to be next broadcast live of selection and content, allow main broadcaster can Periodically to grasp the content that be broadcast live, improve the validity of live streaming according to the update of this list and the list.

Above description system and method class technical solution provided by the present application hereafter recommends to fill to live streaming provided by the present application It sets and is explained.

The live streaming recommendation apparatus provided by the embodiments of the present application applied to server-side is shown referring to Figure 16, Figure 16, is such as schemed Shown in 16, which includes:

Selecting module 1601, for selecting main broadcaster to be recommended for user；Preferably, the selecting module 1601 is processed Journey can be found in the realization of figure 3 above illustrated embodiment.

Generation module 1602, for generating main broadcaster's recommendation list of the user, main broadcaster's recommendation list is for recording The relevant information of the main broadcaster to be recommended；Preferably, the treatment process of the generation module 1602 can be found in real shown in figure 3 above Apply the realization of example.

Sending module 1603, for sending main broadcaster's recommendation list to spectator client.Preferably, the sending module 1603 treatment process can be found in the realization of figure 3 above illustrated embodiment.

Referring to Figure 17, Figure 17 shows the live streaming recommendation apparatus provided by the embodiments of the present application applied to spectator client, As shown in figure 17, which includes:

Receiving module 1701, for receiving main broadcaster's recommendation list of server-side transmission, main broadcaster's recommendation list is for remembering Record the relevant information of main broadcaster to be recommended；Preferably, the treatment process of the receiving module 1701 can be found in and implement shown in figure 6 above The realization of example.

Display module 1702, for being that user shows live streaming relevant information according to main broadcaster's recommendation list.Preferably, should The treatment process of receiving module 1702 can be found in the realization of figure 6 above illustrated embodiment.

The live streaming recommendation apparatus provided by the embodiments of the present application applied to main broadcaster's client is shown referring to Figure 18, Figure 18, As shown in figure 18, which includes:

Receiving module 1801, for receiving the commodity classification recommendation list of server-side transmission, the commodity classification recommends column Table is used to record the end article classification of the user group of concern main broadcaster；The end article classification of the user group is according to institute State what user group obtained the historical operation behavioural analysis of commodity；Preferably, the treatment process of the receiving module 1801 can join See above the realization of embodiment illustrated in fig. 13.

Display module 1802, for being that main broadcaster shows commodity classification according to the commodity classification recommendation list.Preferably, should The treatment process of receiving module 1802 can be found in the realization of figure 13 above illustrated embodiment.

A kind of structure of main broadcaster's recommended models training device provided by the embodiments of the present application is shown referring to Figure 19, Figure 19 Figure, as shown in figure 19, which includes:

Module 1901 is obtained, for obtaining the image comprising characteristics of human body of main broadcaster；Preferably, the acquisition module 1901 Treatment process can be found in the realization of figure 4 above illustrated embodiment.

Detection module 1902 obtains the key point of characterization organization of human body feature for detecting described image；Preferably, the inspection The treatment process for surveying module 1902 can be found in the realization of figure 4 above illustrated embodiment.

Computing module 1903 calculates the corresponding feature vector of each main broadcaster according to the corresponding key point of each image；It is preferred that Ground, the treatment process of the computing module 1903 can be found in the realization of figure 4 above illustrated embodiment.

Generation module 1904, for generating main broadcaster's recommended models according to the feature vector of all main broadcasters.Preferably, the generation The treatment process of module 1904 can be found in the realization of figure 4 above illustrated embodiment.

It should be noted that all the embodiments in this specification are described in a progressive manner, each embodiment weight Point explanation is the difference from other embodiments, and the same or similar parts between the embodiments can be referred to each other. For device class embodiment, system class embodiment, since it is basically similar to the method embodiment, so the comparison of description is simple Single, the relevent part can refer to the partial explaination of embodiments of method.

Finally, it is to be noted that, herein, relational terms such as first and second and the like be used merely to by One entity or operation are distinguished with another entity or operation, without necessarily requiring or implying these entities or operation Between there are any actual relationship or orders.Moreover, the terms "include", "comprise" or its any other variant meaning Covering non-exclusive inclusion, so that the process, method, article or equipment for including a series of elements not only includes that A little elements, but also including other elements that are not explicitly listed, or further include for this process, method, article or The intrinsic element of equipment.In the absence of more restrictions, the element limited by sentence "including a ...", is not arranged Except there is also other identical elements in the process, method, article or apparatus that includes the element.

A kind of live streaming recommended method provided herein, device and system are described in detail above, herein Applying specific case, the principle and implementation of this application are described, and the explanation of above example is only intended to help Understand the present processes and its core concept；At the same time, for those skilled in the art, according to the thought of the application, There will be changes in the specific implementation manner and application range, in conclusion the content of the present specification should not be construed as to this The limitation of application.

Claims

1. a kind of live streaming recommender system characterized by comprising

Server-side generates main broadcaster's recommendation list of the user, the main broadcaster recommends for selecting main broadcaster to be recommended for user List is used to record the relevant information of the main broadcaster to be recommended, sends main broadcaster's recommendation list to spectator client；

Spectator client, the main broadcaster's recommendation list sent for receiving the server-side, according to main broadcaster's recommendation list Live streaming relevant information is shown for user.

2. system according to claim 1, which is characterized in that the server-side is also used to generate commodity classification for main broadcaster Recommendation list sends the commodity classification recommendation list to main broadcaster's client；

Then the system also includes:

Main broadcaster's client is pushed away for receiving the commodity classification recommendation list according to the commodity classification recommendation list for main broadcaster Recommend commodity classification.

3. a kind of live streaming recommended method, which is characterized in that the method is applied to server-side, which comprises

Main broadcaster to be recommended is selected for user；

Main broadcaster's recommendation list of the user is generated, main broadcaster's recommendation list is used to record the correlation of the main broadcaster to be recommended Information；

Main broadcaster's recommendation list is sent to spectator client.

4. according to the method described in claim 3, it is characterized in that,

It is that user selects main broadcaster to be recommended according to the characteristics of human body of all main broadcasters in live data library.

5. according to the method described in claim 4, it is characterized in that,

It is that user selects main broadcaster to be recommended according to main broadcaster's recommended models；Main broadcaster's recommended models are will be in live data library The mathematical model that the characteristics of human body of main broadcaster obtains as training data training.

6. according to the method described in claim 5, it is characterized in that,

Main broadcaster's recommended models are clustered by feature vector of the clustering algorithm to each main broadcaster in live data library What analysis generated, wherein described eigenvector is the multi-C vector for characterizing human face feature.

7. according to the method described in claim 5, it is characterized in that, described to be recommended for user's selection according to main broadcaster's recommended models Main broadcaster, comprising:

8. according to the method described in claim 5, it is characterized in that, described to be recommended for user's selection according to main broadcaster's recommended models Main broadcaster, comprising:

According to main broadcaster's recommended models, the similarity distance between other main broadcasters and the first kind main broadcaster is calculated；It is described other Main broadcaster refers to the main broadcaster removed except the first kind main broadcaster in all main broadcasters；

According to the size relation of similarity distance, the main broadcaster of predetermined number is selected from other described main broadcasters, by selected main broadcaster As main broadcaster to be recommended.

9. according to the method described in claim 5, it is characterized in that, the method also includes:

Commodity classification recommendation list is generated for main broadcaster, sends the commodity classification recommendation list to main broadcaster's client；The commodity Classification recommendation list is used to record the end article classification for the user group for paying close attention to the main broadcaster；The target quotient of the user group Category mesh is to be obtained according to the user group to the historical operation behavioural analysis of commodity.

10. according to the method described in claim 9, it is characterized in that, described generate commodity classification recommendation list, packet for main broadcaster It includes:

For each user in the user group for paying close attention to the main broadcaster, each user merchandise news of interest is obtained, to institute It states merchandise news and carries out the classification of commodity classification；

Institute is calculated according to the interest value of commodity to the interest value of commodity to the historical behavior counting user of commodity according to each user State the interest value of each commodity classification in commodity classification；

According to the interest value sequence size relation of commodity classification, commodity classification recommendation list is generated.

11. a kind of live streaming recommended method, which is characterized in that the method is applied to spectator client, which comprises

Main broadcaster's recommendation list that server-side is sent is received, main broadcaster's recommendation list is used to record the related letter of main broadcaster to be recommended Breath；

12. according to the method for claim 11, which is characterized in that it is described according to main broadcaster's recommendation list be user show Relevant information is broadcast live, comprising:

It is that user shows main broadcaster's information to be recommended, main broadcaster's information to be recommended, comprising: main broadcaster according to main broadcaster's recommendation list One of title, network linking of the face image of main broadcaster, the live video of main broadcaster's publication, main broadcaster's correlation live video are more Kind information.

13. according to the method for claim 11, which is characterized in that it is described according to main broadcaster's recommendation list be user show Relevant information is broadcast live, comprising:

Recommend the relevant information of the main broadcaster of page presentation selection in live streaming, the relevant information of the main broadcaster includes: the face of main broadcaster The network linking of image live video related to main broadcaster.

14. according to the method for claim 11, which is characterized in that it is described according to main broadcaster's recommendation list be user show Relevant information is broadcast live, comprising:

According to main broadcaster's relevant information in the live streaming recommendation list, the relevant live streaming view of main broadcaster is obtained from live data library Frequently；

15. a kind of live streaming recommended method, which is characterized in that the method is applied to main broadcaster's client, which comprises

The commodity classification recommendation list that server-side is sent is received, the commodity classification recommendation list is used to record the use of concern main broadcaster The end article classification of family group；The end article classification of the user group is the history according to the user group to commodity What operation behavior was analyzed；

16. a kind of live streaming recommendation apparatus, which is characterized in that described device is applied to server-side, and described device includes:

Selecting module, for selecting main broadcaster to be recommended for user；

Generation module, for generating main broadcaster's recommendation list of the user, main broadcaster's recommendation list is described wait push away for recording The relevant information of the main broadcaster recommended；

17. a kind of live streaming recommendation apparatus, which is characterized in that described device is applied to spectator client, and described device includes:

Receiving module, for receiving main broadcaster's recommendation list of server-side transmission, main broadcaster's recommendation list is to be recommended for recording Main broadcaster relevant information；

18. a kind of live streaming recommendation apparatus, which is characterized in that described device is applied to main broadcaster's client, and described device includes:

Receiving module, for receiving the commodity classification recommendation list of server-side transmission, the commodity classification recommendation list is for remembering The end article classification of the user group of record concern main broadcaster；The end article classification of the user group is according to the user group Body obtains the historical operation behavioural analysis of commodity；

19. a kind of main broadcaster's recommended models training method, which is characterized in that the method:

20. according to the method for claim 19, which is characterized in that

Described image is the face image comprising face feature.

21. according to the method for claim 20, which is characterized in that it is special that the detection described image obtains characterization organization of human body The key point of sign, comprising:

The key point of characterization face characteristic is obtained using face critical point detection algorithm detection image.

22. according to the method for claim 19, which is characterized in that described to generate main broadcaster according to the feature vector of all main broadcasters Recommended models, comprising:

Clustering is carried out to the corresponding feature vector of all main broadcasters using clustering algorithm and obtains feature clustering matrix, is denoted as Main broadcaster's recommended models.

23. a kind of main broadcaster's recommended models training device, which is characterized in that described device includes:

Module is obtained, for obtaining the image comprising characteristics of human body of main broadcaster；